Proteins that bind small molecules can act as therapeutics by sequestering ligands, stimulating signal- ing pathways, delivering other molecules to sites of action, and serving as in vivo diagnostics. Although the computational design of proteins that can bind to any given ligand is not yet possible, recent successes in de novo enzyme design suggests that it is within reach. Current methods still fail to predict optimal amino acids even in the first shell around the ligand. Short-range interactions with partial covalent character (e.g. hydrogen bonds, salt bridges, and cation--interactions) are often critical for achieving precise positioning within the binding site but are difficult to model becaue their strength is determined by the geometry, polarity, and polarizability of orbitals attached to the interacting functional groups. Existing docking techniques have difficulty handling flexibility of the binding partners, so the structural plasticity of the interface is not taken into account. W believe that computational de novo design of protein-ligand interfaces can not only expand our under- standing of the basic forces involved in molecular recognition, but can also contribute to the development of protein therapeutics, if certain technological limitations can be overcome. The objective of this proposal is to develop a computational protocol for the de novo design of protein- ligand interfaces. The computational design program, ROSETTA, will be expanded through a new scoring func- tion that uses Knowledge-Based Potentials that capture Partial Covalent Interactions (PCI-KBP) at protein- ligand interfaces. Additionally, a fragment-based approach for sampling small molecule conformations will be implemented. The new sampling strategy models ligand flexibility at the binding interface and exploits the speed of amino acid rotamer sampling used for protein design. The accuracy of the computational models will be assessed through redesign and experimental characterization of a panel of 16 protein mutants, each optimized to bind one small molecule out of a focused library of 16 related compounds. Target binding will be determined for the entire set of 16x16=256 combinations using nuclear magnetic resonance (NMR)-based screening experiments. NMR allows detection of weak binding, determination of binding affinities, and verification of the binding site at atomic-level detail. This approach creates a detailed map of the designed interfaces and captures effects on binding through chemical modification of the ligand (derivatization) as well as the protein (mutation). The matrix of experimentally-determined binding affinities will be compared to those predicted by ROSETTA, providing feedback on the accuracy of individual components of the energy function and the efficiency of the sampling strategy. Page: 1